Endoscopists' Acceptance on the Implementation of Artificial Intelligence in Gastrointestinal Endoscopy: Development and Case Analysis of a Scale

Li Tian, Zinan Zhang, Yu Long, Anliu Tang, Minzi Deng, Xiuyan Long, Ning Fang, Xiaoyu Yu, Xixian Ruan, Jianing Qiu, Xiaoyan Wang, Haijun DengFrontiers in Medicine 2022 · DOI

TL;DR

Researchers have developed and validated the first standardized scale to measure how endoscopists perceive and accept Artificial Intelligence (AI) in gastrointestinal procedures. The study reveals that while nearly 80% of endoscopists are optimistic about AI’s accuracy and service benefits, significant concerns remain regarding psychological comfort and the risk of over-dependence on the technology.

What Problem Does This Paper Address?

Artificial Intelligence has shown immense promise in gastrointestinal (GI) endoscopy, with systems like WISENSE and ENDOANGEL helping doctors spot tiny polyps and reduce "blind spots" during procedures. However, technology is only as effective as the people using it.

Until now, there was no validated, standardized way to measure whether endoscopists actually accept these tools. If doctors feel that AI threatens their professional development, creates legal ambiguity in cases of misdiagnosis, or makes them "lazy" (dependent), they may resist its implementation. This paper fills that gap by creating a psychometric scale to quantify these attitudes, helping developers and hospital administrators design better integration strategies.

Methods

The authors followed a multi-stage psychometric development process:

  1. Provisional Scale: Based on literature reviews and the Positive and Negative Affect Schedule (PANAS), they created a 29-item questionnaire.
  2. Refinement: After pilot testing with endoscopists and consulting AI experts, they narrowed the tool down to a 15-item formal scale.
  3. Factor Analysis: They used Confirmatory Factor Analysis (CFA) to group these 15 items into four distinct dimensions (factors):
  4. Service: Impact on workload and patient confidence.
  5. Psychology and Emotion: Interest and comfort levels.
  6. Dependence: The risk of neglecting skills or missing lesions if AI is unavailable.
  7. Accuracy: Perception of the AI’s ability to identify lesions correctly.
  8. Validation: The team tested the scale's reliability (Cronbach’s alpha) and used a Receiver Operating Characteristic (ROC) curve to determine a "cutoff score" that distinguishes between high and low acceptance.

Datasets

The study did not use a traditional machine learning dataset. Instead, it relied on a primary survey dataset:

  • Participants: 210 valid responses from endoscopists across 10 endoscopy centers in China who had prior experience using AI systems.
  • Demographics: Included a mix of ages, professional titles (physicians to chief physicians), and varying levels of experience with AI (from <1 month to >1 year).

Key Insights and Ideas

  • The "Seniority" Paradox: Contrary to the common assumption that younger "digital natives" adopt tech faster, this study found that older endoscopists and those with higher professional titles (Chief Physicians) had higher acceptance scores. The authors suggest senior doctors may view AI as a valuable safety net, whereas juniors might feel it interferes with their fundamental skill acquisition.
  • The Dependence Hurdle: "Dependence" was the lowest-scoring factor. Endoscopists are genuinely worried that relying on AI will cause them to lose their "clinical edge" or neglect to improve their own diagnostic performance.
  • The 50-Point Threshold: Through ROC analysis, the researchers determined that a score of 50 out of 75 serves as a critical threshold. Doctors scoring above 50 are significantly more likely to be satisfied and accepting of AI in their daily practice.
  • Ethics vs. Utility: Interestingly, "Ethics" was a major factor in the early provisional scale but was dropped in the final version because it didn't statistically correlate well with the other factors in the Chinese context. This suggests that for these users, the practical utility and psychological impact of AI currently outweigh abstract legal or ethical concerns.

Key Results

  • High Reliability: The scale achieved an overall Cronbach’s alpha of 0.904, indicating excellent internal consistency.
  • Strong Validity: The correlation between the scale's total score and actual user satisfaction was 0.876, and the Area Under the Curve (AUC) for the ROC was 0.949, showing the scale is highly effective at predicting user acceptance.
  • General Optimism: 79.52% of the surveyed endoscopists fell into the "high acceptance" category.
  • Factor Performance: "Accuracy" scored the highest (mean 4.03/5), while "Dependence" scored the lowest (mean 3.45/5), highlighting exactly where the "trust gap" lies.

Limitations and Future Work

  • Geographic and Institutional Bias: The study was conducted entirely in China, primarily in first-class hospitals. Attitudes might differ in rural settings or different healthcare cultures (e.g., Western Europe or North America).
  • Novelty Effect: Because AI is relatively new in GI endoscopy, the high acceptance scores might be influenced by a "honeymoon phase." Long-term studies are needed to see if acceptance wanes as the novelty wears off.
  • Scale Evolution: The authors acknowledge that as AI becomes more integrated, the scale may need to be updated to include more specific items regarding data privacy and the evolving legal landscape of medical AI.

Summary updated

Abstract

This study yielded a viable questionnaire to measure the acceptance among endoscopists of the implementation of AI in gastroenterology.
BibTeX
@inproceedings{tian2022endoscopists,
  author = {Li Tian and Zinan Zhang and Yu Long and Anliu Tang and Minzi Deng and Xiuyan Long and Ning Fang and Xiaoyu Yu and Xixian Ruan and Jianing Qiu and Xiaoyan Wang and Haijun Deng},
  title = {{Endoscopists' Acceptance on the Implementation of Artificial Intelligence in Gastrointestinal Endoscopy: Development and Case Analysis of a Scale}},
  booktitle = {Frontiers in Medicine},
  year = {2022},
  doi = {10.3389/fmed.2022.760634},
}
Figure